Greetings from Tattle!

In April it seemed hard to imagine that any topic could overtake coronavirus in the misinformation storm raging around us. But 2020 is the year of possibilities! In June 2020, Bollywood actor Sushant Singh Rajput was found dead in his house in Mumbai. In the months that followed, we saw conspiracy theories relating to the actors death overtake Covid conspiracies by many miles. In line with prior research findings, misinformation on social media closely mirrored topics on mainstream television news.

Amid Covid and the crazy news and 'fake news' cycles, Tattle has had the busiest couple of months. In our first year we spent time putting together the building blocks that we thought were necessary for any intervention, be it research or technical, around misinformation in India. Tattle started by creating datasets that help understand the misinformation trends in India- only 17 million Indians use Twitter. More than double that number use regional apps such as Helo and Sharechat. There is a need for datasets beyond Twitter that give a more representative understanding of information trends and behaviours evolving in India.

Quick Updates

• We now have a corpus of over 700,000 unique social media posts in Indian languages.
• We released the Khoj App Beta.
• We were honored to host a stellar panel on what GPT-3 means for machine learning in Indian languages.
• We released a dataset paper on claims in user generated content.
• We were interviewed for a piece on security on WhatsApp- Are private chats secure?.

Tattle Blog and Talks:

• Tech Talk on Machine Learning for multi-lingual, multi-modal content
• The Unexplained Motivation Behind WhatsApp Forwards
• Deepfakes, Cheap Fakes and Misinformation in India

Data for the Win

In addition to crawling fact-checking sites and making that data searchable, we have been archiving content from regional apps for six months. We now have a fast-growing corpus of over 700,000 unique social media posts in four Indian languages. Data collected by Tattle has been used by several researchers and journalists over the last few months. To highlight a few:

• Misinformation as a Window into Prejudice: COVID-19 and the Information Environment in India. CSCW, 2020
• The strange, sometimes sinister conspiracy theories on Sushant Singh Rajput's death that flourished on social media. The Economic Times, October 2020.
• Coronavirus: The Human Cost of Fake News in India. BBC, June 2020.

We have been finding ways of making the collected data more digestible for news consumers as well as researchers. We released a dashboard that allows exploration of broad themes in misinformation in a week.

We also released a paper from our annotation of posts from Indian social media platforms. Some insights from our research:

• Over 60% of posts on social media contain a factual claim. This goes to show that fact checking all social media content would be a very daunting task.
• 'Selfie videos' are the most common kind of videos shared online.
• Majority of content doesn't contain an identifiable source.

During the literature review we were surprised at the scant attention paid to multi-media or user generated content, commonly found on WhatsApp, in prior literature on misinformation detection. This research gap needs urgent attention if we are to cope with misinformation in vast parts of the world.

Work in This Quarter

A lot of foundational work still remains to be done- building Indian language specific technologies is a continuing research and engineering challenge. We are however, now entering the second phase of our work where we are building products on the foundations established in our first year. The beta for the Khoj App is out. The app leverages the fact-checking sites database to help users share accurate information on WhatsApp. We are currently doing a pilot study to understand what interventions would work well for the 'Ruhi Masis' of India.

What’s next?

• Opening up the pilot study to 1000 users and on-boarding them to the Khoj App.
• Making the archive UI more accessible for people to search and navigate the data
• Expanding the scope of our multimedia Search engine
• Building privacy preserving tools that let users share their WhatsApp histories with the archive.
• More partnerships with researchers working on social media and misinformation in the Global South.

How Can You be Involved?

• Join the Community Response team for the Khoj App.
• If you know a regional Indian language, help us understand the accuracy of AI tools in that language.
• If you’re a developer, head over to GitHub to help out with data science and software development.
• If you know of funding opportunities aligned with Tattle's mission, do send them our way.
• Share this project with your networks so the community can grow!

Shout Out to Collaborators!

A big thank you to all these people who’ve supported us with their time:

• Anushree Abhyankar for working on video search.
• Nikhil Bararia for helping us identify content in Bengali and Rajasthani that we should archive.
• Rishabh Jain for supporting user research and implementation of the Khoj App.
• Shubham Atreja for being our sounding board for research ideas.
• Sudeep Duggal for improving the fact checking scraper.

Stay in Touch

You can follow us on Twitter, join our slack channel or this mailing list for more structured project updates.

For specific questions, feel free to email us at admin@tattle.co.in